Summary: This enables appending PKG_CONFIG_PATH to env variable

Reviewed By: wez

Differential Revision: D17363236

fbshipit-source-id: 18c6d7a97ba83edf085278bccaafa80821ea8860

Summary:
This commit configures CI for Windows, macOS and Linux using the
new GitHub Actions CI service.

The configuration was generated via:

```
build/fbcode_builder/getdeps.py generate-github-actions --output-file .github/workflows/main.yml folly
```
Pull Request resolved: https://github.com/facebook/folly/pull/1228

Reviewed By: yfeldblum

Differential Revision: D17454557

Pulled By: wez

fbshipit-source-id: a99261ee312d4093031cf6fad5832eacb1be1d97

Summary: [Folly] Cut no-op same-type casts in `Function` benchmark.

Reviewed By: LeeHowes

Differential Revision: D17432433

fbshipit-source-id: 6c2e53fc742aa590f7332efa146af41592274cc9

Summary: Original commit changeset: e28b4cb055b3

Reviewed By: yfeldblum

Differential Revision: D17429976

fbshipit-source-id: e2f6f59c22854f1b2fd25913ad779f4edc83f8ee

Summary:
create a unorderedReduceSemiFuture method that returns a SemiFuture
instead of just a Future.

Reviewed By: LeeHowes

Differential Revision: D16946582

fbshipit-source-id: e28b4cb055b3a9c67adebb373c60a775f7005762

Summary:
Update the thrift C++ and Python CMake rules to indicate that the output also
depends on the thrift compiler itself.

Previously the C++ rule indicated that the output depended on the thrift
template files, which caught most cases when the thrift compiler was updated,
but wasn't fully correct.  The thrift templates were also removed and baked
into the thrift compiler binary in D16356056.

Reviewed By: yfeldblum, chadaustin

Differential Revision: D17401217

fbshipit-source-id: ae5cde7a7e5e07a74406a1b6f4469124187bc12f

Summary:
This is useful when working on changes to some of the builder functions,
to skip ever trying to use cached results.

Reviewed By: chadaustin

Differential Revision: D17401219

fbshipit-source-id: fb7d5ea45618653957b9bd6a62eed91d8334824d

Summary:
Changes libfolly.pc to a more standard format.

Allows overriding of the variables by pkgconfig.
Signed-off-by: Rosen Penev <rosenp@gmail.com>

I'm not sure of anything that uses pkgconfig to find libfolly but might as well fix it.
Pull Request resolved: https://github.com/facebook/folly/pull/1226

Reviewed By: simpkins

Differential Revision: D17415620

Pulled By: yfeldblum

fbshipit-source-id: d015c3e83783b1c5e9ac48a64ad35c0c07120dbd

Summary:
This commit teaches fixup-dyn-deps how to generate correct
absolute paths in the context of the ultimate install path (specified
via the `--final-install-prefix` option)

Absolute paths are desirable if you have, for example, an executable
that you wish to install with the setuid bit set.

Reviewed By: simpkins

Differential Revision: D17361491

fbshipit-source-id: 4c4f3f15208266300622f84dc9cd1ba83883dfb7

Summary:
Avoids -Wundef on non-Windows builds.
folly/system/ThreadId.h had #elif _WIN32 which was changed to
#elif defined(_WIN32)

Reviewed By: Orvid

Differential Revision: D17399760

fbshipit-source-id: 76583374c33d5d59e7c64b9a7a05c660a817bb92

Summary:
Add a license header to satisfy the open source linter. Use the same
header the other .cmake files have.

Reviewed By: mhlakhani

Differential Revision: D17404782

fbshipit-source-id: 66679d72c9e680f8bb8b27869e981a046b3520cf

Summary:
This commit adds a getdeps command that is able to generate
a workflow file for the GitHub Actions CI environment.

The workflow file could be expressed more simply using the matrix
syntax and with three steps (checkout, build, test), but I chose to
break out the steps for each of the dependencies because the UX
while waiting on the build is much nicer that way: the steps show
during and live log tailing for the section of the build that is
underway.  If they were all lumped into a single build step then
the logs from the boost section of the build dominate and make
the github UI work very hard.

Pull Request resolved: https://github.com/facebook/watchman/pull/743

Test Plan:
https://github.com/facebook/watchman/pull/743 successfully
executes the github actions CI flow.

```
$ opensource/fbcode_builder/getdeps.py generate-github-actions --output-file watchman/.github/workflows/main.yml watchman
```

Reviewed By: simpkins

Differential Revision: D17384915

Pulled By: wez

fbshipit-source-id: 9a9e5a3e806c18f6cc38ba1cb7059740cda01ad4

Summary:
GitHub Actions CI `windows-latest` environment has only VS 2019
installed, so we need to expand our logic to be able to locate it.

Note that Boost 1.69 doesn't know how to locate VS 2019 so we are effectively
tied to VS 2017 at the moment; the search order in this diff reflects that.

(That means that we can't target `windows-latest` on GitHub Actions, but that
is really a concern for a later diff in this stack)

Reviewed By: simpkins

Differential Revision: D17385052

fbshipit-source-id: 9bb0612154f42d425a625406488f39bb4ec3d8ae

Summary:
while testing https://github.com/facebook/watchman/pull/743 I
noticed that the cmake builds were picking up the installed mingw GCC
compiler rather than the MSVC compiler.  That would be fine except that
boost is built with MSVC and its generated libraries cannot be subsequently
found by a cmake gcc build that uses FindBoost.

This commit forces cmake to pick cl.exe rather than gcc.  This is probably
fine to do unconditionally on windows, but since I've only observed this
particular problem with GitHub Actions I'm keeping it constrained to that
environment for now.

Reviewed By: simpkins

Differential Revision: D17385050

fbshipit-source-id: 90bef898b138e5d4bbd28a155ed1bd468acee9de

Summary:
We've been squeaking by with assuming that flex is installed already
on posix systems, but that isn't the case on the github actions default
configuration.

Adjust the bison recipe: on windows it deploys both flex and bison.  We use the
same source for both flex and bison but install flex to a separate install
prefix to make it easier to consume the flex dependency distinct from the bison
dependency.

The latest flex release segfaults during compilation on linux unless we
force -DGNU_SOURCE, so the manifest does that on linux.

Reviewed By: simpkins

Differential Revision: D17385051

fbshipit-source-id: 9f31b07849af9de50099d1b20bedba517bbbdf2f

Summary:
while testing https://github.com/facebook/watchman/pull/743 on macOS
I noticed that the libevent build failed to find openssl.

openssl is special on macos because apple do not ship the headers.
We already build and depend on openssl for the folly manifest on
macos, so this is really just adding a missing edge.

Reviewed By: simpkins

Differential Revision: D17385053

fbshipit-source-id: 1b688537fef422d81a959fc5749c871b9e868baa

Summary: Diffing logs would be easier if the lines which pretty-printed JSON were identical if their values were identical, and they're just easier to visually scan for a value if they're sorted anyway.

Reviewed By: yfeldblum, luciang

Differential Revision: D17367669

fbshipit-source-id: 91188f23576ce39b9edc2307be018a43f30c786d

Summary: [Folly] Use a fixed mempool capacity in `LifoSem`, simplifying usage.

Reviewed By: nbronson

Differential Revision: D17251982

fbshipit-source-id: e3bce99204daef178418268f1d195365400df885

Summary: AsyncGeneratorWrapper uses folly::Optional as the return type for "getNext". Now the AsyncGenerator API has changed by adding `next()` method which returns a `NextResult` type, which works like folly::Optional while supporting both values and references, it's best to use `NextResult` for the wrapper as well.

Reviewed By: yfeldblum

Differential Revision: D16695661

fbshipit-source-id: 84e9d74e8d6b17b93bd3c22e2aaf56c7b9e4b789

Summary:
We would like to build a version of proxygen that has minimal
dependencies on dynamic libraries.

Reviewed By: yfeldblum

Differential Revision: D17228181

fbshipit-source-id: cfd61afdfa978c49a536184f426502196241fb8a

Summary: broken zinc/test and tao/coordinator/test: in dev/nosan mode.

Reviewed By: luciang

Differential Revision: D17310046

fbshipit-source-id: 0a98d7cea24438a42932466394e63db4f64591cf

Summary:
On Windows we have to explicitly invoke `make_fbpy_archive.py` with python.
Therefore use CMake's built-in `FindPythonInterp` module to find the python
executable and use that when invoking `make_fbpy_archive.py`

This is slightly complicated by the effort required to find python with older
versions of CMake.  We ideally still want to support versions of CMake back to
at least 3.8, which means we need to still support finding Python with the
older `FindPythonInterp.cmake` module

Reviewed By: wez

Differential Revision: D17128606

fbshipit-source-id: 3f4beff76848b8a362ebdf21198e7a8bf1b0537f

Summary: The nested EBO classes were missing the allocator-extended c'ctor.  Fix and add tests to cover this case.

Reviewed By: nbronson

Differential Revision: D17284711

fbshipit-source-id: 6e7e8df83712f018fd6a7558513164f319b1b7f8

Summary: [Folly] Cut unused dep on boost `chrono` library from folly and proxygen.

Reviewed By: vitaut

Differential Revision: D17298687

fbshipit-source-id: 90d0182f4fbb76c13cb341e27514e688f12590a4

Summary: [Folly] Move `hash_128_to_64` to top of file - it has no dependencies so there was no need to split the declaration and definition across 500 lines.

Reviewed By: Orvid

Differential Revision: D17276906

fbshipit-source-id: f57133e8873d8b1afcfd5ab119e07615f6ee9018

Summary:
Basic idea is to find the subprogram for a given stack trace address, scan the inlined subroutines of that subprogram recursively for all inline function calls, then find the name and location of these inline function calls.

Now only pure inline functions are supported, and class inline functions are not. Class inline functions needs a lookup path "caller->obj->type", then lookup types in .debug_types, which is more complex.

The debug info parser idea was stolen from hphp/tools/debug-parser/dwarfstate.h

Not sure if we want to add a flag to enable this feature. We can discuss over review.

Reviewed By: luciang

Differential Revision: D16760775

fbshipit-source-id: 1a7c23486605e7d8837070a697a875789ae41416

Summary: This adds a file-level comment that explains the various usages.

Differential Revision: D16259169

fbshipit-source-id: 95e5e8f047cdd159899cab47c390a7b0ab8d70cb

Summary:
gcc 9's libstdc++ moves the definition of _M_impl._M_finish
into a new intermediate class. It's still addressible with the same name,
but the type of the member pointer changes. This diff switches to using
decltype to tolerate both the old and the new implementation.

Reported as issue https://github.com/facebook/folly/issues/1223

Reviewed By: yfeldblum

Differential Revision: D17283175

fbshipit-source-id: 9c9da28b83c3cb3584d33575b462d10b17002edb

Summary: [Folly] No mutable flags in IOBuf to avoid races when invoking `const`-qualified members concurrently.

Reviewed By: simpkins

Differential Revision: D17251909

fbshipit-source-id: e52d333fb620d05f7229a11db9c2f14fe9f6f028

Summary:
[Folly] Avoid ambiguous call in futures. Because `std::atomic` is in `namespace std`, unqualified calls to `atomic_compare_exchange_strong_explicit` from within `namespace folly` are ADL-ambiguous between the overloads in `namespace std` and in `namespace folly`.

Also:
* Fix incorrectly-named prototype.
* Fix used of `std::atomic<T>::value_type` which may not exist.

Fixes #1221.

Reviewed By: aary, Orvid

Differential Revision: D17256763

fbshipit-source-id: ffcaf171e6257cdef29de0ffb63f310d7f4a93eb

Summary: Making a new ctor so that we can pass in the parameters

Reviewed By: yfeldblum

Differential Revision: D17173662

fbshipit-source-id: ab518892694e1bec1179177e7d9477e3a89ff38d

Summary:
Compiling folly on Arch Linux triggered a few platform-specific behaviors that were not covered correctly by tests.

- Fix F14{Map,Set} maxSize test cases
    The maximum size of `F14Table` is dependent of its allocator's max_size:

    ```
    max_size() -> std::min<std::size_t>(
        (std::numeric_limits<InternalSizeType>::max)(),
        AllocTraits::max_size(a)
    )
    ```

    This commit fixes the `F14{Map,Set}` maxSize tests, so they won't fail
    on platforms where the standard allocator's `max_size()` is based on
    `ptrdiff_t` (signed).

-  Fix TestUtilTest
    If glog is compiled without gflags support, `SetCommandLineOptionWithMode()`
    can't force glog to use stderr.

    glog has its own compatibility layer for `FLAGS_*` in case gflags is missing,
    that's why `FLAGS_logtostderr` is always available.

- Skip AsyncIOTest tests if O_DIRECT is not supported by the file system
    `fs::temp_directory_path()` returns `"/tmp"` on Linux.

    When using systemd, the /tmp directory is mounted by default as tmpfs;
    and tmpfs does not support the `O_DIRECT` flag.

    This commit skips tests in case the file can't be opened with `O_DIRECT`.
Pull Request resolved: https://github.com/facebook/folly/pull/1218

Reviewed By: nbronson

Differential Revision: D17160597

Pulled By: yfeldblum

fbshipit-source-id: 00274d1f8770079e35559ee9bd8426402e949abd

Summary: [Folly] Switch from `-defs.h` pattern to `-inl.h` pattern in `folly/stats/` since `extern template` doing shallow instantiation rather than deep instantiation makes this pattern difficult.

Reviewed By: chadaustin

Differential Revision: D16761319

fbshipit-source-id: 1760bf53b09791bebf996064c0d1fcc3babf43d2

Summary: [Thrift] Bake templates into the compiler binary rather than having to package the templates with the compiler.

Reviewed By: stevegury, vitaut

Differential Revision: D16356056

fbshipit-source-id: b971dce7aada8dc2a26587fbf818e661acd0c555

Summary:
[Folly] Extract compare-exchange hack under TSAN to a common location and make its interface mimic `std::atomic_compare_exchange_strong_explicit`. Limit it to Clang TSAN (no knowledge of whether it applies to other implementations of TSAN), but extend it to handle all pairs of success and failure orders.

The TSAN bug is described in https://github.com/google/sanitizers/issues/970. High level, Clang TSAN ignores the explicit failure order and infers a failure order from the success order - which is broken if the explicit failure order is in the relevant sense stronger than the success order.

Reviewed By: aary, nbronson

Differential Revision: D16854217

fbshipit-source-id: 18f6458520bbd5f482e41c10d7229e6cfae1db2a

Summary:
This should make it easier to eg: enable fbzmq on macos at a later
time, and also makes things more explicit about what is being built, and
importantly, by listing these in the manifest ensures that we have a
hash change if we change this list; we wouldn't trigger such a change
if the list were encoded solely in builder.py

Reviewed By: chadaustin

Differential Revision: D17133149

fbshipit-source-id: caf0dd45e262188eeefafe0868ef95ad257a4950

Summary:
* Fix a minor issue with `add_fb_python_executable` where it would
  emit `--main` even if `MAIN_MODULE` had not been specified
* Introduce the ability to rename python sources in the manifest so
  that they have a different install path.
* Note that this changes the DEPENDS parameters to reference abs_sources;
  previously these were using relative paths to the sources pulled directly
  from the SOURCES arguments, but since those can now contain aliasing
  operations we need to pre-process the list.  I switched to abs_sources
  because that was already being maintained.  If for some reason the
  DEPENDS list needs to be relative paths, this will need to be adjusted.

Reviewed By: simpkins

Differential Revision: D17147417

fbshipit-source-id: d0825bfbdad6be658c078aaac6645e6926a1530f

Summary:
This diff adds a small abstraction that allows for uploading
and downloading from an artifact cache.

We try to download from this cache at build time, but will only
try to populate it for continuous builds--those are built from
code that has been reviewed and landed on master.  This restriction
helps to avoid thrashing the cache with works in progress and
results in a slightly more trustworthy state of the cache contents.

In addition to this, we choose only to cache third party projects.
The rationale is that our first party projects move too quickly to
be worth caching, especially since the cache granularity is for
the whole project rather than just changed elements of a given
project.

In a later diff I will introduce some implementations of the
cache class that work with eg: Travis or Circle CI caching.

Reviewed By: simpkins

Differential Revision: D16873307

fbshipit-source-id: 2bfb69e36615791747b499073586562f2ca48be9

Summary: one_ms is short enough that it can trigger under testing. Increase to make less likely to fail spuriously.

Reviewed By: yfeldblum

Differential Revision: D17190402

fbshipit-source-id: 091cd3ee8f3da224e39ba95c403e233115a80510

Summary: SemiFuture::within was undocumented and Future::within's use of the term "throws" was confusing. This cleans up both sets of documentation.

Reviewed By: yfeldblum

Differential Revision: D17186392

fbshipit-source-id: 83409df85b07bd3cad9fb020e99fc3a7f3c70b61